How datastax/metric-collector-for-apache-cassandra works
Project type and purpose
• This is an open-source monitoring agent/tool, not a full-stack web app.
• It consists of:
• A Java “agent” jar that you attach to Cassandra nodes
• Configuration templates & scripts for collectd
• A built-in Prometheus exporter endpoint
• Dashboards (Grafana + Prometheus) with demo deployment
File structure
• src/main/java/...: Core Java agent code
• com.datastax.mcac.Agent – entry point loaded by the JVM
• CollectdController, UnixSocketClient – send metrics/events to collectd
• interceptors/insights/metrics utils – transform Cassandra metrics & diagnostic events
• config/: default YAML and collectd.conf templates
• dashboards/:
• demo/ – docker-compose for quick start
• grafana/ – jsonnet templates & generated dashboards
• prometheus/ – prometheus.yaml + target list
• k8s-build/ – Helm-style templating for Kubernetes
• scripts/datalog-parser.py – post-processing of on-node JSON logs
Key components
A. Cassandra Node
• Cassandra JVM
• MCAC Java agent jar (injected via -javaagent)
• Local disk for “datalog” files (optional)
• Metric interceptors (CassandraDaemonInterceptor)
• Insight/event generator
• FilteringRule & ConfigurationLoader
• UnixSocketClient → Collectd
• Embedded Prometheus exporter (HTTP server on port 9103)
C. Collectd daemon (external process on same host)
• Uses config/collectd.conf.tmpl + plugin set
• Gathers OS metrics + receives metrics over local Unix socket
D. Monitoring Backend
• Prometheus server (scrapes MCAC exporter & Collectd if configured)
• Grafana (dashboards from dashboards/grafana)
• Docker-Compose or k8s resources that stand up:
• Cassandra cluster containers (with agent)
• Prometheus container
• Grafana container
F. Optional: Datalog parser script (post-agent JSON logs → human analysis / jq)
Relationships & data flow
1. Cassandra JVM emits metrics/events → MCAC Agent interceptors
2. MCAC Agent formats metrics/events → writes to local collectd via Unix socket
3. Collectd collects OS + C* metrics → (optional) pushes to other backends OR via Prometheus plugin
4. MCAC Agent’s embedded HTTP exporter → Prometheus scrapes on port 9103
5. Prometheus stores time series → Grafana queries Prometheus
6. Grafana renders dashboards
Architectural patterns / design principles
• Agent-based instrumentation (drop-in JVM agent)
• Adapter layer: Cassandra metrics → collectd format → Prometheus metrics
• Plug-in driven (collectd plugins, jsonnet dashboards, k8s templates)
• Config-driven (YAML for filtering rules, collectd, dashboards)
Technologies / frameworks to call out
• Java agent (javaagent API)
• collectd (with Unix socket plugin)
• Prometheus exporter library (embedded in agent)
• Grafana + jsonnet for dashboard generation
• Docker‐Compose / Kubernetes manifests for deployment
• Python script for datalog parsing